1 Experiments

  • I am moving to a more longer scale, semester like personal cycle. Each cycle is of 3-4 months and is pretty similar to OKRs. I am seeing a few positive changes in what I am able to do now but since there are other confounding factors, I won't say anything very clear at the moment.
  • Gave a rough try to speaker embeddings from Real-Time-Voice-Cloning models for a tiny internal dataset. Even though the accents are pretty different (US trained, tested on Indian English), the embeddings are pretty informative and useful.
  • I am trying to see how useful I really can make speech/acoustics as a mode of interacting with my machine(s). I have a live list here.

2 Readings/Explorations

3 Programming

Commits for week 40-2019 and 4 previous weeks.

Little less than usual commit counts is an interesting thing to notice. These last couple of weeks had me doing the sort of programming I like rather than what I am forced to do usually. For earlier weeks, I believe I was still doing regular stuff but just a little less.

4 Media

In a world which could not be grasped as a whole, and where there were no universally shared values, most people clung to the particular niche to which they were most committed: their job or profession. They treated their work as a post-religious calling, ‘an absolute end in itself’, and if the modern ‘ethic’ or ‘spirit’ had an ultimate foundation, this was it.


  • [breck2016s] Breck, Cai, Nielsen, Salib & Sculley. 2016. "What’s your ML Test Score? A rubric for ML production systems." , , link. doi.
  • [jia2018transfer] Jia, Zhang, Weiss, Wang, Shen, Ren, Nguyen, Pang, Moreno, Wu & others. 2018. "Transfer learning from speaker verification to multispeaker text-to-speech synthesis", 4480-4490, in in: Advances in neural information processing systems, edited by
  • [carlini2019secret] Carlini, Liu, Erlingsson, Kos & Song. 2019. "The Secret Sharer: Evaluating and testing unintended memorization in neural networks", 267-284, in in: 28th $\$USENIX$\$ Security Symposium ($\$USENIX$\$ Security 19), edited by
  • [wan2018generalized] Wan, Wang, Papir & Moreno. 2018. "Generalized end-to-end loss for speaker verification", 4879-4883, in in: 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), edited by
  • [mignan2019one] Mignan & Broccardo. 2019. "One neuron is more informative than a deep neural network for aftershock pattern forecasting." arXiv preprint arXiv:1904.01983, , link. doi.
  • [le2014distributed] Le & Mikolov. 2014. "Distributed representations of sentences and documents", 1188-1196, in in: International conference on machine learning, edited by